It has been known for some time that conventional direct methods can serve as a valuable tool for locating the positions of heavy-atom substructures using either isomorphous or anomalous difference structure factors. Note that SnB applications to anomalous difference data DO NOT necessarily result in a substructure with the correct hand. It is the magnitude of the difference that is used. Therefore, the probability that substructure coordinates from an SnB solution will have the correct hand is 50%.

Most of the successful applications of SnB to SeMet data sets have involved the use of peak-wavelength anomalous difference |E |s, and experience has shown that these applications are highly dependent on the accuracy of the difference magnitudes. Therefore, since the amount of data available for substructure problems is relatively large, the user should be stringent in eliminating all data resulting from uncertain measurements. Since the probability of very large difference |E |s (e.g. >5.0) is remote, data sets that appear to have many such measurements should be examined for measurement errors. If such data persist even after the adoption of rigorous rejection criteria, they should be eliminated using the maximum |E| parameter on the reflections & invariants screen.

Parameters. Substructure applications of SnB to difference data are basically very similar to full structure applications. Users should pay attention to the following:

1. On the General Information screen
   1. Select either SIR or SAS data type
   2. Fill in the contents line with the substructure formula
   3. Enter the anomalous dispersion information for SAS data
2. On the Create Es screen
   1. For SIR data, enter the required derivative information
   2. Supply the ASU contents information for the complete molecule
3. Make certain that SnB has chosen the Fourier Grid Size appropriately

The parameter recommendations and default values for difference-data applications to substructures differ slightly from those for complete structures.

• Given Nu heavy (substructure) atoms:
• Phases: 20Nu-40Nu (default 30Nu)
• Invariants: 200Nu-400Nu (default 300Nu)
• Peaks recycled/picked: 0.8*Nu-Nu (default Nu)
• Shake-and-Bake cycles: 2Nu

Selecting Correct Sites. The higher the density of a peak on the E-map, the higher the probability that the site is real. Typically, there is no demarkation that marks the separation between good and bad peaks, and it is possible for some spurious peaks to appear in higher density than some correct peaks. The visualization feature can be used to check for close interpeak distances that may make certain peaks questionable. If it appears that more than one trial is a solution, cross correlation between the peak lists can be helpful because correct peaks will occur in more than one trial, but spurious peaks often do not. This comparison must be done carefully, taking into account all possible origins and both enantiomorphs, and the option to do it within SnB will be added in the future. A prudent approach is to take the best 70-80% of the peaks and compute a difference map, checking for reoccurrence of the lower ranking peaks.

Reflection/Invariant Problems. When using difference data, it is relatively common for SnB to encounter problems satisfying the requested number of reflections and triplet invariants, and an error message appears in a pop-up window. To correct these problems and proceed with the job, the following parameter values must be adjusted on the reflections & invariants screen: (1) the number of reflections, (2) the number of triplet invariants, and (3) the |E|/sig(|E|) ratio (Zmin). A shortage of reflections may occur because too few reflections satisfy the difference-magnitude significance cut-offs. Failure to generate a sufficient number of invariants occurs because the percentage of the total available reflections being used for phasing is much less than it is in the full-structure case, and the Miller indices are spread over a much larger range so that the reflections simply do not interact as strongly to form triplets. This can normally be overcome by increasing the number of reflections used since the number of invariants that can be generated tends to increase exponentially. Thus, the user must attempt to optimize the relevant parameters while keeping the following in mind:

• try to use the minimum number of most reliable reflections but never reduce the atom:phase ratio to less than 1:10,
• try not to reduce Zmin to less than 2.0, and
• do not use a phase:invariant ratio less than 1:5.

In bad cases, it is necessary to rerun DREAR while relaxing the other difference cut-offs. In the worst cases, it is necessary to collect more accurate data.

Using Fa Values. If Fa values have been derived from MAD data, the Basic data-type option can also be used to locate the sites for an anomalously scattering substructure. If Fa values are being used, DO NOT apply the Bayesian correction to the weak intensities, but DO increase the numbers of reflections and invariants to use by 5-6 times the default values for Basic applications.